OpenAI's ChatGPT Work Turns ChatGPT Into an Autonomous Agent That Does the Job Across Every App

OpenAI is trying to move ChatGPT past the chat box. Front and center at the company's global Build Week — a seven-day, 60-plus-event developer blitz across 49 cities — is ChatGPT Work, an autonomous agent, powered by the new GPT-5.6 family, that gathers context from a user's connected apps and files and returns finished work: documents, spreadsheets, presentations, and interactive dashboards. Launched on July 9 and now being showcased to developers worldwide, it reframes ChatGPT from something you prompt into something that does the job while you're away.

The pitch is autonomy across surfaces. ChatGPT Work can execute multi-step tasks across web, phone, and desktop, run in the background, browse the web through a built-in browser, and reach into your local files and apps — connected through plugins that wire your systems into the agent. Rather than handing back a wall of text, it produces the actual deliverable: a formatted document, a multi-sheet spreadsheet with working formulas, a slide deck, or a small interactive web app for a report or dashboard. OpenAI even highlights improved "design judgment," pitching the output as tasteful and functional rather than just correct.

The most consequential plumbing is on the desktop. OpenAI has merged its Codex coding app into a new ChatGPT desktop app for Mac and Windows, with the previous app renamed "ChatGPT Classic." That means Codex — OpenAI's coding agent and the star of Build Week — now ships inside ChatGPT itself, so the same surface that writes your quarterly deck can also read a codebase, fix a bug, and stand up a web app. Existing Codex users keep their projects, settings, and workflows when they update. It's a deliberate collapse of "coding tool" and "work assistant" into one agent.

Under the hood sits GPT-5.6, released the same day in three tiers: Sol, the flagship for the heaviest work; Terra, the balanced everyday model; and Luna, the cost-efficient option for free and lower tiers. ChatGPT Work rolled out first to Pro, Enterprise, and Edu plans, with Plus and Business following within days. OpenAI's framing is unusually direct about the goal: GPT-5.6 Sol is billed as doing more with fewer tokens — roughly 54% more token-efficient on agentic coding — which is exactly what makes an always-on background agent economically viable.

Strategically, ChatGPT Work is OpenAI's answer to a market that has decided the future of AI is agents, not chatbots. Anthropic has been pushing Claude Cowork into marketing, legal, and finance teams; Google is wiring Gemini agents across Workspace. By folding Codex into ChatGPT and handing hundreds of millions of existing users an agent that produces real work across their apps, OpenAI is betting on distribution: the fastest way to make agentic AI the default isn't a new product people have to discover, but a new capability inside the one they already open every day. Build Week is the invitation to developers to build on top of it.